No member named 'begin' in namespace 'std'

Question

I successfully compiled on Windows a code that should be crossplatform. Now when compiling it in Xcode with Mac OS X, I get:

std::valarray<float> v(32);
...
std::sort(begin(v), end(v));            # Use of undeclared identifier 'begin'
std::sort(std::begin(v), std::end(v));  # No member named 'begin' in namespace 'std'
std::sort(std::valarray::begin(v), std::valarray::end(v));  # Idem, error as well

Why does the error No member named 'begin' in namespace 'std' happen?

Non-member `begin` is a C++11 feature, try enabling it in XCode. — TartanLlama, Oct 05 '16 at 13:16
@Basj: just saw that while testing :) I confirm that adding -std=c++11 to the command line it compiles fine. I'm glad I haven't answered without testing :) — Jean-François Fabre, Oct 05 '16 at 13:18
when will std=c++11 will become default? come on it's 2016 and there's a lot of questions only solved by adding that switch! — Jean-François Fabre, Oct 05 '16 at 13:19
@Jean-FrançoisFabre I assume it is some old XCode verson? I mean, it wouldn't be user-friendly to not include C++11, and this is on a mac, axiomatically it is user-friendly. heh. — Yakk - Adam Nevraumont, Oct 05 '16 at 13:23
Just for anyone else looking up, here's one more possibility with maps. If your map is a data member and you're trying to access it inside a non-static class method that is qualified as `const`, then accessing mymap[x] will result in compile time error, because, mymap[x] will create an entry for x if it's already not there, so it's modifying the data member in a function that's marked as `const`, so violation. Use mymap.at(x), if you know that x is already present in mymap, using map.find(x) — ppadhy, Sep 22 '21 at 06:21

score 6 · Answer 1 · edited May 23 '17 at 11:55

6

std::begin was introduced with C++11. See this answer for how to enable C++11 in XCode 4.2 (the precise name of the dialect has probably changed by now).

Alternatively, if you can't move to C++11, switch to std::vector and use v.begin().

edited May 23 '17 at 11:55

Community

1
1

answered Oct 05 '16 at 13:20

Martin Bonner supports Monica

28,528
3
51
88

Thanks. About your alternate solution, I get `v.begin()` : `No member named 'begin' in 'std::valarray'` – Basj Oct 05 '16 at 13:21
@Basj: Yeah. Sorry about that - I saw your comment above and updated my alternative. – Martin Bonner supports Monica Oct 05 '16 at 13:23
1

Even if the OP has C++11 `valarray` does not have a begin and end member function. – NathanOliver Oct 05 '16 at 13:24
@NathanOliver: Yes - which is why I suggested switching to vector and using v.begin if he can't get `begin(v)` to work. – Martin Bonner supports Monica Oct 05 '16 at 13:26
@NathanOliver: Following on from your comment on a deleted post: is defined to provide an overload of `std::begin` for `valarray`s. – Martin Bonner supports Monica Oct 05 '16 at 13:27
@MartinBonner Good call. Missed it was added – NathanOliver Oct 05 '16 at 13:28
@NathanOliver: Turning on C++11 *is* all you have to do to get the code to work. – Martin Bonner supports Monica Oct 05 '16 at 13:28
*Somewhere* there should be a "C++ Language Dialect" option - you need to change that (and probably "C++ Standard Library" too). – Martin Bonner supports Monica Oct 05 '16 at 13:40
@MartinBonner In fact, I cannot use C++11 because it requires OS 10.7 SDK. And for legacy support (my software needs it), I have to use Base SDK = OS X 10.5. When I enable C++ Language Dialect and C++ Standard Library as you explained, I get `clang error invalid deployment target for stdlib libc++ (requires OS 10.7 or later)` – Basj Oct 05 '16 at 20:03
Ah. In that case, either switch to vector + v.begin(), or use &v[0], (&v[0])+v.size(). I'm not sure valarray is guaranteed to be contiguous, but in practise it will be.. – Martin Bonner supports Monica Oct 05 '16 at 20:24
@MartinBonner [It seems to be contiguous](http://stackoverflow.com/a/11143184/1422096). What is the difference between your `&v[0]` and @Yakk's `T* begin( std::valarray& v ) { return addressof(v[0]); }` ? In which case the latter is better? – Basj Oct 05 '16 at 20:54
The difference is that Yakk has taken the trouble to factor out the "get the start of the container" into a separate function. – Martin Bonner supports Monica Oct 05 '16 at 20:59
@MartinBonner In practice, are there equivalent? I admit I don't understand what does `T* addressof(T& arg) { return reinterpret_cast( &const_cast( reinterpret_cast(arg))); }` do. Too much cast for me! – Basj Oct 05 '16 at 21:01
The problem with &v[0] is that it may not work if the type has an operator&() defined. addressof is guaranteed to work, and all the casts are to keep the compiler quiet. – Martin Bonner supports Monica Oct 05 '16 at 21:03

score 2 · Answer 2 · edited May 23 '17 at 10:34

2

You could be compiling in C++03 mode. Work out how to get your IDE to compile in C++11 mode. XCode 4.2 turn on C++11 may help.

std::sort(std::valarray::begin(v), std::valarray::end(v)); -- I don't think the standard demands this ever work. I guess if valarray implemented begin and end as statics or Koenig friend operators or somesuch.

std::valarray doesn't come with member begin/end. The only way in C++03 to iterate over it was to use [] or with pointers.

valarray is guaranteed to be contiguous.. So you can write

namespace notstd {
  // addressof taken from http://en.cppreference.com/w/cpp/memory/addressof
  template<class T>
  T* addressof(T& arg) {
    return reinterpret_cast<T*>(
           &const_cast<char&>(
              reinterpret_cast<const volatile char&>(arg)));
  }

  template<class T>
  T* begin( std::valarray<T>& v ) { return addressof(v[0]); }
  template<class T>
  T* end( std::valarray<T>& v ) { return begin(v)+v.size(); }
  template<class T>
  T const* begin( std::valarray<T> const& v ) { return addressof(v[0]); }
  template<class T>
  T const* end( std::valarray<T> const& v ) { return begin(v)+v.size(); }
}

and use notstd::begin on your valarray.

edited May 23 '17 at 10:34

Community

1
1

answered Oct 05 '16 at 13:21

Yakk - Adam Nevraumont

262,606
27
330
524

`begin` and `end` are guaranteed to be in: , , , , , , , , , , and .; as well – NathanOliver Oct 05 '16 at 13:23
1

@NathanOliver: ... and which is the one of interest here! – Martin Bonner supports Monica Oct 05 '16 at 13:24
@MartinBonner That because `valarray` does not have a `begin` or `end` member function. It would be silly to include it as you cannot use it. By the looks of it any standard header that has begin/end has begin/end included which is a reasonable thing to do. – NathanOliver Oct 05 '16 at 13:26
Thanks @Yakk. In which case this code (with adressof...) is better than replacing begin by `&v[0]` ? – Basj Oct 05 '16 at 20:55
The real win is in the definition of `end` (in practise `addressof` is hardly ever needed - it's more important for authors of library code that has to work with arbitrary types). – Martin Bonner supports Monica Oct 05 '16 at 20:58
@basj Martin has it. – Yakk - Adam Nevraumont Oct 05 '16 at 21:38

No member named 'begin' in namespace 'std'

2 Answers2

Linked

Related